Chefkoch Search Recipes by Diet

Purpose

Search Chefkoch.de's catalog (≈384,000 recipes) for those matching one or more diet/allergen filters and optionally a keyword, returning each recipe's title, canonical URL, image, rating + vote count, preparation time, difficulty, short description, and recipe ID. Also covers the daily "Was koche ich heute?" suggestion stream when no diet filter is requested. Read-only — never posts, rates, saves to Kochbuch, or interacts with the Wochenplaner.

When to Use

• "Find me vegan main-course recipes rated ≥4 stars."
• "Give me 30-minute low-carb dinners."
• "What's on Chefkoch's daily 'Was koche ich heute' picks?"
• "Glutenfreie Suppe für 4 Personen" (free-text keyword + diet tag).
• Bulk extraction of search-result metadata for meal-planning agents, recipe aggregators, or RAG pipelines targeting German-language cooking content.

Workflow

Chefkoch's search-result page is fully server-rendered into a stable URL grammar with embedded data-vars-* attributes on every recipe card and a parallel <script type="application/ld+json"> ItemList. There is no JSON API, no GraphQL, and no XHR endpoint needed — browse cloud fetch <url> --proxies returns the full result set in one HTTP request (~0.5s for 30-result page, status 200, no anti-bot stealth required). Lead with the URL-param HTTP path; reserve the browser flow only for inspecting the filter modals interactively, which the agent shouldn't need to do at runtime.

1. Construct the search URL

URL grammar (all modifiers optional except s<page>):

https://www.chefkoch.de/rs/s<page>[t<id>,<id>,…][r<minRating>][p<maxMinutes>][o<sort>][/<query>]/<slug>-Rezepte.html

• s<N> — page index, zero-based. s0 = first page (≤30 cards), s1 = next page, etc. The trailing -Rezepte.html slug is informational only — the server canonicalizes whatever you submit, so you can pass /Rezepte.html and follow the 301 to learn the canonical slug. A request with s<N> beyond the available page count 301-redirects back to s1 (verified s30 → s1 for vegan). Read Location header to detect exhaustion.
• t<id>[,<id>,…] — comma-separated tag IDs from the table below. The server auto-sorts ascending and rewrites the URL via 301 (e.g. t57,21 → t21,57). Combining tags ANDs them: t21,57 returns recipes that are both Hauptspeise AND Vegan.
• r<n> — minimum average rating, 1..5. Decimal allowed (r4.5). r4 returned 18,491 vegan recipes vs. 26,393 unfiltered.
• p<minutes> — maximum preparation time ("Arbeitszeit") in literal minutes. Verified values: p15, p30, p60, p120. The UI slider in the modal exposes 5 stops (15/30/60/120/unlimited) but the URL accepts any positive integer.
• o<n> — sort order: o2 Empfehlung (default — can be omitted), o3 Bewertung (best-rated first), o6 Neuheiten (newest first). Other values 301 back to default.
• /<query>/ — optional free-text keyword between the modifier block and the slug, e.g. /rs/s0t57/glutenfrei/Rezepte.html for vegan + "glutenfrei" keyword. Use this for allergens/ingredients not in the diet enum (gluten-free, laktosefrei, nussfrei, sojafrei, etc.) — Chefkoch's diet enum doesn't cover specific allergens beyond Vegan/Vegetarisch, so allergy intent falls back to keyword.

Tag ID reference (complete enum, extracted 2026-05-20 from /rs/s0/Rezepte.html)

Diet (Ernährung) — primary group for this skill:

Meal type (Mahlzeit): 21 Hauptspeise · 19 Vorspeise · 36 Beilage · 90 Dessert · 71 Snack · 53 Frühstück

Category (Rezeptkategorie): 30 Auflauf · 82 Pizza · 94 Reis-/Nudelsalat · 15 Salat · 3669 Salatdressing · 122 Tarte · 52 Fingerfood · 35 Dips · 34 Saucen · 40 Suppe · 166 Klöße · 108 Brot/Brötchen · 46 Brotspeise · 51 Aufstrich · 89 Süßspeise · 127 Eis · 92 Kuchen · 147 Kekse · 93 Torte · 157 Confiserie · 11 Getränke · 113 Shake · 313 Gewürzmischung · 243 Pasten · 211 Studentenküche

Characteristic (Rezepteigenschaften): 50 Einfach · 49 Schnell · 79 Basisrezepte · 48 Preiswert

Cuisine (Länderküche): 65 Deutsch · 28 Italienisch · 43 Spanisch · 149 Portugiesisch · 84 Französisch · 117 Englisch · 86 Osteuropäisch · 133 Skandinavisch · 44 Griechisch · 103 Türkisch · 212 Russisch · 163 Naher Osten · 14 Asiatisch · 13 Indisch · 148 Japanisch · 38 Amerikanisch · 74 Mexikanisch · 95 Karibisch · 114 Lateinamerikanisch · 101 Afrikanisch · 131 Marokkanisch · 168 Ägyptisch · 145 Australisch

Occasion (Anlass): 78 Frühling · 27 Sommer · 99 Herbst · 64 Winter · 91 Für Kinder · 120 Ostern · 39 Halloween · 102 Weihnachten · 87 Silvester · 106 Festlich · 63 Grillen · 152 Camping · 45 Party

Technique (Zubereitung): 144 Kochen · 69 Braten · 107 Dünsten · 153 Blanchieren · 66 Schmoren · 23 Backen · 137 Überbacken · 116 Wok · 135 Mikrowelle · 162 Römertopf · 62 Fondue · 136 Marinieren · 134 Frittieren · 128 Flambieren · 150 Haltbarmachen · 1147 Wursten

Example URLs (all return HTTP 200, fully rendered HTML with 30 recipe cards on s0):

# Vegan, sorted by rating, min 4 stars
https://www.chefkoch.de/rs/s0t57r4o3/Vegan-Rezepte.html      # 18,491 results

# Low-carb main courses, min 4 stars, sorted by rating
https://www.chefkoch.de/rs/s0t21,9948r4o3/Hauptspeise-Low-Carb-Rezepte.html

# Vegetarian + glutenfrei keyword, max 30 minutes
https://www.chefkoch.de/rs/s0t32p30/glutenfrei/Vegetarisch-Rezepte.html

# Vegan pasta dishes
https://www.chefkoch.de/rs/s0t57/pasta/Vegan-Rezepte.html    # 2,767 results

2. Fetch the page

browse cloud fetch "<URL>" --proxies > /tmp/page.json

--proxies flag is optional (Chefkoch has no anti-bot enforcement against direct US IPs in our testing) but cheap insurance against future enforcement and adds <100ms latency. No --verified stealth needed. Response is wrapped in a JSON envelope — the HTML is at .content.

3. Parse the total result count + recipe cards

import re, json, html

# Total count (German number format, '.' as thousands separator)
m = re.search(r'(\d[\d.]+)\s*Rezepte', content)
total = int(m.group(1).replace('.', '')) if m else None

# Each recipe card has the same shape — extract by data-vars-* attributes
card_re = re.compile(
    r'data-vars-position="(?P<pos>\d+)"\s+'
    r'data-vars-tracking-id="recipe-(?P<id>\d+)"\s+'
    r'data-vars-recipe-title="(?P<title>[^"]+)"'
    r'.{0,200}?data-vars-num-votes="(?P<votes>\d+)"\s+'
    r'data-vars-rating="(?P<rating>[\d.]+)"\s+'
    r'data-vars-has-video="(?P<video>[01])"',
    re.DOTALL,
)

# Within each card, time + difficulty live in <div class="ds-recipe-info">
# with two <span class="ds-recipe-info__text"> children:
#   1st: "10 Min." | "1 Std." | "2 Std. 15 Min."
#   2nd: "simpel" | "normal" | "pfiffig"
info_re = re.compile(
    r'<div class="ds-recipe-info">.+?'
    r'<span class="ds-recipe-info__text"[^>]*>([^<]+)</span>.+?'
    r'<span class="ds-recipe-info__text"[^>]*>([^<]+)</span>',
    re.DOTALL,
)

# Canonical URL is the first <a class="ds-recipe-card__link"> href
url_re = re.compile(r'<a class="ds-recipe-card__link[^"]*"\s+href="([^"]+)"')

# Description (1-line teaser, optional)
desc_re = re.compile(r'<div class="ds-recipe-card__description[^"]*">([^<]+)</div>')

# Image (full-size 640x800 lives in srcset)
img_re = re.compile(r'<img alt="[^"]+"\s+loading="lazy"\s+src="([^"]+)"')

The JSON-LD ItemList block (<script type="application/ld+json">) is a parallel, lighter-weight surface that gives you {position, url, name, description} per recipe without HTML scraping — use it when you only need title + URL and want to skip data-vars regex. Cross-check: both surfaces return 42 items per s0 page on filtered queries, 30 on others. (The page-DOM data-vars-position count and the JSON-LD count match exactly per response.)

4. Paginate

Increment s<N> until the response 301-redirects you back to s1 (signaling exhaustion) or you hit your target N:

def fetch_page(url): ...
page = 0
while True:
    url = f"https://www.chefkoch.de/rs/s{page}t57r4o3/Vegan-Rezepte.html"
    r = fetch_page(url)
    if r['statusCode'] == 301 and '/rs/s1' in r['headers']['Location']:
        break  # exhausted
    cards = parse(r['content'])
    if not cards: break
    yield from cards
    page += 1

Page size is ~30 cards (sometimes 42 with embedded "you-might-also-like" tail). Plan request budget around ceil(total / 30) pages.

5. Daily "Was koche ich heute?" (no filter)

When the user has no diet/keyword intent and just wants today's editorial picks:

browse cloud fetch "https://www.chefkoch.de/rezepte/was-koche-ich-heute/" --proxies

Returns 8 hand-curated recipes for the day. Extract them from the page's recipe URLs (https://www.chefkoch.de/rezepte/<id>/<slug>.html) — the JSON-LD on this page is a CollectionPage, not an ItemList, so recipe URLs surface via plain anchor regex rather than schema.org structure. The same card shape (data-vars-rating, data-vars-num-votes, etc.) is not present on this page; titles come from the anchor's title= attribute or surrounding <h3>.

Browser fallback

Only needed if you want to (a) trigger fresh-Vue rendering of a non-public filter the URL grammar doesn't expose, or (b) demo the diet-modal UI to a human. The runtime skill should not need this. If you do open a browser:

sid=$(browse cloud sessions create --keep-alive | node -e "...parse .id...")
export BROWSE_SESSION="$sid"
browse open "https://www.chefkoch.de/rs/s0/Rezepte.html" --remote
# Dismiss the cookie consent modal — required before any interaction
browse snapshot --remote                            # find button: "Allen Zwecken zustimmen"
browse click @<ref-from-snapshot> --remote
# Click filter chip "Ernährung", check diet boxes, click "Anwenden".
# Read the URL the page navigates to — it is exactly the URL grammar above.
browse cloud sessions update "$sid" --status REQUEST_RELEASE

The cookie consent dialog (Willkommen bei Chefkoch — Allen Zwecken zustimmen) blocks all interaction until dismissed; it does not appear on browse cloud fetch requests because no cookies are persisted, so it is irrelevant for the recommended URL-param path.

Site-Specific Gotchas

• No anti-bot enforcement on direct HTTP fetches in 2026-05-20 testing — bare curl + User-Agent would likely work, and browse cloud fetch --proxies certainly works (status 200, no captcha, no Akamai). Don't waste time on --verified stealth sessions for this skill.
• Server canonicalizes tag order and rewrites URL: submitting t57,21 → 301 to t21,57/Hauptspeise-Vegan-Rezepte.html. Always follow the redirect chain; the final URL is the canonical one Chefkoch indexes. browse cloud fetch does not auto-follow — read the Location header and re-fetch.
• Slug at end is informational only — /Vegan-Rezepte.html, /Hauptspeise-Vegan-Rezepte.html, even /Rezepte.html all work for the same tag set. The server uses the modifier block to compute its own slug and 301s if you submit a "wrong" one. Don't try to reconstruct the slug from tag names; just submit /Rezepte.html and let the server rewrite.
• Pagination beyond the result set 301s to s1 rather than serving an empty page. Use this as a clean exhaustion signal: if you sent s<N> and the response is 301 Location: /rs/s1..., you've gone past the last page.
• No dedicated allergen filter beyond Vegan/Vegetarisch: the diet enum (10 values) covers vegetarian, vegan, low-carb, ketogen, paleo, low-fat, low-cal, Trennkost (food-combining), Vollwert (whole-foods), and Baby — but does not include glutenfrei, laktosefrei, nussfrei, sojafrei, fructosearm, or histaminarm. For those, fall back to free-text keyword search via the /<query>/ segment (e.g. /rs/s0t32p30/glutenfrei/Rezepte.html). The keyword matches recipe titles, descriptions, ingredient lists, and tags.
• r<n> rating is a minimum threshold, not a tier — r4 means ≥4.00, not "4-star tier". Decimal precision works (r4.5 → 10,505 vegan results vs. r4 → 18,491).
• o2 sort is the default and can be omitted — submitting o2 explicitly will 301-redirect to a URL without the o modifier. Only o3 (rating) and o6 (newest) survive in canonical URLs.
• p<minutes> is in literal minutes, not the slider's 1–5 index. The UI slider snaps to 15/30/60/120; the URL accepts any positive integer.
• German thousands separator in result count: "26.393 Rezepte" means 26,393, not 26.393. Strip the period before parsing.
• Card count per page varies: filtered queries return ~42 cards/page (30 results + 12 cross-sell at tail); broad queries return 30. The data-vars-position attribute is the authoritative item index — trust it over len(matches).
• JSON-LD ItemList is lossy — it carries {position, url, name, description} but no rating, vote count, prep time, difficulty, or image. Use the data-vars regex on the card HTML for full metadata, or hit each recipe page individually for richer schema.org Recipe data.
• Difficulty enum is fixed: simpel, normal, pfiffig. Not localized in the URL.
• Cooking-time strings are not zero-padded — "5 Min.", "45 Min.", "1 Std.", "1 Std. 30 Min.". Parse as duration; don't assume two-digit minutes.
• "Was koche ich heute?" page does NOT carry data-vars-* on cards and uses a CollectionPage JSON-LD type, not ItemList — its 8 daily recipes need plain-anchor extraction. Use a different parser than the search-result parser for this page.
• Cookie consent (dialog: Willkommen bei Chefkoch) blocks the browser UI — only relevant if you open a browser session. The "Allen Zwecken zustimmen" button (extracted ref e.g. [5-1009]) is what dismisses it. browse cloud fetch is unaffected.
• PLUS-paywalled recipes exist as a small fraction of results, marked by data-vars-plus-content="1" (and the "Nur PLUS-Rezepte" filter toggle). They surface in the result list but the recipe detail page is gated behind a subscription. The skill should still return them as listed; downstream consumers can filter on the is_plus flag.

Expected Output

For a typical filtered query (/rs/s0t57r4o3/Vegan-Rezepte.html):

{
  "query": {
    "url": "https://www.chefkoch.de/rs/s0t57r4o3/Vegan-Rezepte.html",
    "tags": [{"id": 57, "label": "Vegan", "group": "diet"}],
    "min_rating": 4.0,
    "max_prep_minutes": null,
    "sort": "rating",
    "keyword": null,
    "page": 0
  },
  "total_results": 18491,
  "page_size": 30,
  "recipes": [
    {
      "position": 1,
      "id": "4410071768223128",
      "title": "Veganes Gnocchi-Ragout",
      "url": "https://www.chefkoch.de/rezepte/4410071768223128/Veganes-Gnocchi-Ragout.html",
      "image": "https://img.chefkoch-cdn.de/rezepte/4410071768223128/bilder/1614973/crop-640x800/veganes-gnocchi-ragout.jpg",
      "rating": 4.71,
      "num_votes": 7,
      "prep_time_minutes": 10,
      "prep_time_raw": "10 Min.",
      "difficulty": "normal",
      "description": "vegan",
      "has_video": true,
      "is_plus": false
    }
  ]
}

For the daily-suggestions page (/rezepte/was-koche-ich-heute/):

{
  "query": {"url": "https://www.chefkoch.de/rezepte/was-koche-ich-heute/", "type": "daily_suggestions"},
  "date": "2026-05-20",
  "total_results": 8,
  "recipes": [
    {
      "position": 1,
      "id": "111751046790147",
      "title": "Käsesuppe",
      "url": "https://www.chefkoch.de/rezepte/111751046790147/Kaesesuppe.html",
      "rating": null,
      "num_votes": null,
      "prep_time_minutes": null,
      "difficulty": null
    }
  ]
}

For an exhausted pagination request (s<N> past last page):

{
  "query": {"url": "https://www.chefkoch.de/rs/s99t57/Vegan-Rezepte.html", "page": 99},
  "total_results": 18491,
  "page_size": 0,
  "recipes": [],
  "exhausted": true,
  "redirected_to": "https://www.chefkoch.de/rs/s1t57/Vegan-Rezepte.html"
}

For an unknown allergen the diet enum doesn't cover (glutenfrei, laktosefrei, nussfrei), combine a related diet tag with the keyword segment:

{
  "query": {
    "url": "https://www.chefkoch.de/rs/s0t32p30/glutenfrei/Rezepte.html",
    "tags": [{"id": 32, "label": "Vegetarisch", "group": "diet"}],
    "keyword": "glutenfrei",
    "max_prep_minutes": 30,
    "note": "Chefkoch's diet enum has no glutenfrei tag — falling back to keyword match."
  }
}